Conventional methods for assessing levels of sensory-motor impairment in stroke patients are inherently subjective and dependent upon the clinician's own observations and opinions. In this study, 93 control and 63 stroke subjects underwent robotic assessment to gauge sensory-motor impairment. Multiple statistical data normalization and dimensionality-reduction measures were evaluated, using four different classifier types, in order to derive an optimal feature vector for the purpose of distinguishing stroke from control subjects. The optimal feature vector was then utilized to train a committee of classifiers for the purpose of recombining data from several traditional sensory-motor assessment scores into a KINARM specific assessment metric. We were able to create a training vector capable of distinguishing between stroke and control subjects with high accuracy, and demonstrated that the committee of classifiers assigned consistent scores to patients of similar levels of impairment.